I think your taking the processing too far. You shouldn't really think of it as "Noise Removal" but as "noise reduction" Try reducing the bakground instead of removing it entirely. Sometimes echos appear when the processing is pushed to far and other times they were always there but masked by the background noise.
"Noise removal" is a shot in the dark based on statistics. Noise is usually a stochastic, random signal, so it has a non-zero chance of matching the model of speech of the noise removal program. Conversely, speech has a non-zero chance of matching the noise model. Particularly consonants and specifically fricatives are virtually indistinguishable from noise.
So the more "noise removal" you apply, the more your speech gets affected along with the noise.
The main purpose of noise removal with speech is not to restore the original quality of a signal source but to increase its comprehensibility. This comes at the cost of musical noise, echoes and other artifacts.
If you want a low-noise recording for a voice-over, you don't want the distracting qualities of noise removal: you are almost always better off with a constant noise floor of tolerable loudness. So try moving closer to your microphone or getting better recording equipment. You'll likely also want to apply some noise-gating (which is a rather coarse form of noise removal but with less impact on the speech) but of course you'll be better off if you don't have significant noise levels to start with.